Figure 1:

minicore k-means++ is faster than scikit-learn k-means++. Performance evaluation (y-axis) of elapsed time (seconds) for sparse data (top) and log₁₀ transformed time for dense data (bottom) for increasing sizes of k (x-axis) for the PBMC dataset with 68k cells (left), Cao et al. dataset with 2 million cells (middle), and Cao et al. dataset with 4 million cells (right). Results for minicore k-means++ are in red (standard) and green (with localsearch++); scikit-learn k-means++ is blue.